perf: defer imports + runtime micro-optimizations in hot paths

[KRRT7]

Stacked on #12.

E2E Import Time (hyperfine, 30+ runs, Standard_D2s_v5)

CPython 3.12 (typing imports re — deferral doesn't help)

Import	master	+ #12	+ this PR
Console	79.1 ± 0.8ms	37.6 ± 1.3ms	37.5 ± 0.5ms
RichHandler	100.3 ± 3.6ms	39.0 ± 0.5ms	39.6 ± 0.5ms

CPython 3.13 (typing no longer imports re — full benefit)

Import	master	+ #12 (est.)	+ this PR	Speedup vs master
Console	67.9 ± 0.7ms	~36.8ms	33.6 ± 0.5ms	2.02x
RichHandler	—	~38.4ms	37.5 ± 0.4ms	—

On Python ≥3.13, typing no longer imports re, so deferring all re.compile() calls eliminates re (+ _sre, re._compiler, re._parser, re._constants) from the Console import chain entirely, saving ~3ms.

Runtime Micro-benchmarks (Standard_D2s_v5, Python 3.13.13)

Benchmark	Before	After	Speedup
Style.__eq__ (identity)	114ns/call	62ns/call	1.84x
Style.__eq__ (equal)	113ns/call	129ns/call	~same
Style.combine (3 styles)	579ns/call	433ns/call	1.34x
Style.combine (2 styles)	604ns/call	550ns/call	1.10x
Style.chain (3 styles)	959ns/call	878ns/call	1.09x
Segment.simplify (identity)	1269ns/call	931ns/call	1.36x
Segment.simplify (equal)	1269ns/call	1128ns/call	1.12x
E2E Console.print	173.7µs/call	171.6µs/call	~1.01x

Identity comparisons (same object) are the common case in the render pipeline — cached styles are reused across segments. The E2E improvement is modest (~1%) because these hot paths are a fraction of total Console.print cost.

What this adds (on top of #12)

color.py — deferred imports

• from __future__ import annotations
• Defer colorsys, _palettes, terminal_theme to first use in get_truecolor() / downgrade() (both LRU-cached)
• Move Result to TYPE_CHECKING

Defer re module from Console import chain (7 files)

• color.py: RE_COLOR compiled lazily in Color.parse() (LRU-cached)
• text.py: _re_whitespace compiled lazily on first rstrip_end(); inline import re in highlight_regex(), highlight_words(), split(), detect_indentation(), with_indent_guides()
• markup.py: RE_TAGS via _compile_tags(), RE_HANDLER lazy in render(), escape regex lazy in escape(); add from __future__ import annotations, move Match to TYPE_CHECKING
• _emoji_replace.py: regex default arg → lazy _EMOJI_SUB global; move Match to TYPE_CHECKING
• _wrap.py: re_word → lazy _re_word in words()
• highlighter.py: import re moved inside JSONHighlighter.highlight()
• default_styles.py: 3 rgb(...) color strings → Color.from_rgb() to avoid regex-based Color.parse() at import time

Runtime micro-optimizations in hot paths

• style.py Style.__eq__/__ne__: add identity shortcut (is) before hash comparison — skips hash computation when comparing the same cached object (very common in the render pipeline)
• style.py Style.combine/chain: use _add (LRU-cached) directly instead of sum() which routes through __add__ + redundant .copy() check per iteration
• segment.py Segment.simplify: check is before == for style comparison — adjacent segments very often share the exact same Style object reference

Testing

• pytest tests/ — 952 passed, 25 skipped
• Updated tests/test_markup.py to use _compile_tags() instead of removed RE_TAGS module constant